Model Safety

OpenAI has disclosed six newly identified safety issues involving its models, marking a fresh effort to make failures and risks more visible. The incidents were reported since March and appear to span problems significant enough to warrant public documentation, though the available reports do not provide a full breakdown of their technical causes, severity, or affected products. Together, the disclosures highlight the difficulty of evaluating increasingly capable models across many uses and deployment environments.

Alongside the disclosures, OpenAI is launching an incident-tracking system intended to organize and communicate model safety events. Such a system can help researchers, developers, policymakers, and users distinguish isolated glitches from recurring patterns, follow how problems are investigated, and assess whether mitigations reduce risk over time. It also creates a more consistent channel for reporting issues as models evolve. The move places OpenAI’s safety reporting practices under greater public scrutiny while contributing to a broader industry debate over transparency, accountability, and responsible deployment. The cluster focuses on both the six reported cases and the infrastructure OpenAI is creating to monitor similar incidents in the future.